datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
   3  * Distributed under the terms of the GNU GPL version 2.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 /* Functions for managing the dp interface/device. */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/init.h>
  14 #include <linux/module.h>
  15 #include <linux/if_arp.h>
  16 #include <linux/if_vlan.h>
  17 #include <linux/in.h>
  18 #include <linux/ip.h>
  19 #include <linux/jhash.h>
  20 #include <linux/delay.h>
  21 #include <linux/time.h>
  22 #include <linux/etherdevice.h>
  23 #include <linux/genetlink.h>
  24 #include <linux/kernel.h>
  25 #include <linux/kthread.h>
  26 #include <linux/mutex.h>
  27 #include <linux/percpu.h>
  28 #include <linux/rcupdate.h>
  29 #include <linux/tcp.h>
  30 #include <linux/udp.h>
  31 #include <linux/version.h>
  32 #include <linux/ethtool.h>
  33 #include <linux/wait.h>
  34 #include <asm/system.h>
  35 #include <asm/div64.h>
  36 #include <asm/bug.h>
  37 #include <linux/highmem.h>
  38 #include <linux/netfilter_bridge.h>
  39 #include <linux/netfilter_ipv4.h>
  40 #include <linux/inetdevice.h>
  41 #include <linux/list.h>
  42 #include <linux/rculist.h>
  43 #include <linux/dmi.h>
  44 #include <net/inet_ecn.h>
  45 #include <net/genetlink.h>
  46
  47 #include "openvswitch/datapath-protocol.h"
  48 #include "checksum.h"
  49 #include "datapath.h"
  50 #include "actions.h"
  51 #include "flow.h"
  52 #include "vlan.h"
  53 #include "tunnel.h"
  54 #include "vport-internal_dev.h"
  55
  56 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  57     LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
  58 #error Kernels before 2.6.18 or after 3.0 are not supported by this version of Open vSwitch.
  59 #endif
  60
  61 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  62 EXPORT_SYMBOL(dp_ioctl_hook);
  63
  64 /**
  65  * DOC: Locking:
  66  *
  67  * Writes to device state (add/remove datapath, port, set operations on vports,
  68  * etc.) are protected by RTNL.
  69  *
  70  * Writes to other state (flow table modifications, set miscellaneous datapath
  71  * parameters such as drop frags, etc.) are protected by genl_mutex.  The RTNL
  72  * lock nests inside genl_mutex.
  73  *
  74  * Reads are protected by RCU.
  75  *
  76  * There are a few special cases (mostly stats) that have their own
  77  * synchronization but they nest under all of above and don't interact with
  78  * each other.
  79  */
  80
  81 /* Global list of datapaths to enable dumping them all out.
  82  * Protected by genl_mutex.
  83  */
  84 static LIST_HEAD(dps);
  85
  86 static struct vport *new_vport(const struct vport_parms *);
  87 static int queue_userspace_packets(struct datapath *, u32 pid, struct sk_buff *,
  88                                  const struct dp_upcall_info *);
  89
  90 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  91 struct datapath *get_dp(int dp_ifindex)
  92 {
  93         struct datapath *dp = NULL;
  94         struct net_device *dev;
  95
  96         rcu_read_lock();
  97         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
  98         if (dev) {
  99                 struct vport *vport = internal_dev_get_vport(dev);
 100                 if (vport)
 101                         dp = vport->dp;
 102         }
 103         rcu_read_unlock();
 104
 105         return dp;
 106 }
 107 EXPORT_SYMBOL_GPL(get_dp);
 108
 109 /* Must be called with genl_mutex. */
 110 static struct flow_table *get_table_protected(struct datapath *dp)
 111 {
 112         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
 113 }
 114
 115 /* Must be called with rcu_read_lock or RTNL lock. */
 116 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 117 {
 118         return rcu_dereference_rtnl(dp->ports[port_no]);
 119 }
 120
 121 /* Must be called with rcu_read_lock or RTNL lock. */
 122 const char *dp_name(const struct datapath *dp)
 123 {
 124         return vport_get_name(rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]));
 125 }
 126
 127 static int get_dpifindex(struct datapath *dp)
 128 {
 129         struct vport *local;
 130         int ifindex;
 131
 132         rcu_read_lock();
 133
 134         local = get_vport_protected(dp, OVSP_LOCAL);
 135         if (local)
 136                 ifindex = vport_get_ifindex(local);
 137         else
 138                 ifindex = 0;
 139
 140         rcu_read_unlock();
 141
 142         return ifindex;
 143 }
 144
 145 static inline size_t br_nlmsg_size(void)
 146 {
 147         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 148                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 149                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 150                + nla_total_size(4) /* IFLA_MASTER */
 151                + nla_total_size(4) /* IFLA_MTU */
 152                + nla_total_size(1); /* IFLA_OPERSTATE */
 153 }
 154
 155 /* Caller must hold RTNL lock. */
 156 static int dp_fill_ifinfo(struct sk_buff *skb,
 157                           const struct vport *port,
 158                           int event, unsigned int flags)
 159 {
 160         struct datapath *dp = port->dp;
 161         int ifindex = vport_get_ifindex(port);
 162         struct ifinfomsg *hdr;
 163         struct nlmsghdr *nlh;
 164
 165         if (ifindex < 0)
 166                 return ifindex;
 167
 168         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 169         if (nlh == NULL)
 170                 return -EMSGSIZE;
 171
 172         hdr = nlmsg_data(nlh);
 173         hdr->ifi_family = AF_BRIDGE;
 174         hdr->__ifi_pad = 0;
 175         hdr->ifi_type = ARPHRD_ETHER;
 176         hdr->ifi_index = ifindex;
 177         hdr->ifi_flags = vport_get_flags(port);
 178         hdr->ifi_change = 0;
 179
 180         NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
 181         NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
 182         NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 183 #ifdef IFLA_OPERSTATE
 184         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 185                    vport_is_running(port)
 186                         ? vport_get_operstate(port)
 187                         : IF_OPER_DOWN);
 188 #endif
 189
 190         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
 191
 192         return nlmsg_end(skb, nlh);
 193
 194 nla_put_failure:
 195         nlmsg_cancel(skb, nlh);
 196         return -EMSGSIZE;
 197 }
 198
 199 /* Caller must hold RTNL lock. */
 200 static void dp_ifinfo_notify(int event, struct vport *port)
 201 {
 202         struct sk_buff *skb;
 203         int err = -ENOBUFS;
 204
 205         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 206         if (skb == NULL)
 207                 goto errout;
 208
 209         err = dp_fill_ifinfo(skb, port, event, 0);
 210         if (err < 0) {
 211                 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 212                 WARN_ON(err == -EMSGSIZE);
 213                 kfree_skb(skb);
 214                 goto errout;
 215         }
 216         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 217         return;
 218 errout:
 219         if (err < 0)
 220                 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 221 }
 222
 223 static void release_dp(struct kobject *kobj)
 224 {
 225         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 226         kfree(dp);
 227 }
 228
 229 static struct kobj_type dp_ktype = {
 230         .release = release_dp
 231 };
 232
 233 static void destroy_dp_rcu(struct rcu_head *rcu)
 234 {
 235         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 236
 237         flow_tbl_destroy(dp->table);
 238         free_percpu(dp->stats_percpu);
 239         kobject_put(&dp->ifobj);
 240 }
 241
 242 /* Called with RTNL lock and genl_lock. */
 243 static struct vport *new_vport(const struct vport_parms *parms)
 244 {
 245         struct vport *vport;
 246
 247         vport = vport_add(parms);
 248         if (!IS_ERR(vport)) {
 249                 struct datapath *dp = parms->dp;
 250
 251                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 252                 list_add(&vport->node, &dp->port_list);
 253
 254                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 255         }
 256
 257         return vport;
 258 }
 259
 260 /* Called with RTNL lock. */
 261 void dp_detach_port(struct vport *p)
 262 {
 263         ASSERT_RTNL();
 264
 265         if (p->port_no != OVSP_LOCAL)
 266                 dp_sysfs_del_if(p);
 267         dp_ifinfo_notify(RTM_DELLINK, p);
 268
 269         /* First drop references to device. */
 270         list_del(&p->node);
 271         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 272
 273         /* Then destroy it. */
 274         vport_del(p);
 275 }
 276
 277 /* Must be called with rcu_read_lock. */
 278 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 279 {
 280         struct datapath *dp = p->dp;
 281         struct sw_flow *flow;
 282         struct dp_stats_percpu *stats;
 283         int stats_counter_off;
 284         int error;
 285
 286         OVS_CB(skb)->vport = p;
 287
 288         if (!OVS_CB(skb)->flow) {
 289                 struct sw_flow_key key;
 290                 int key_len;
 291                 bool is_frag;
 292
 293                 /* Extract flow from 'skb' into 'key'. */
 294                 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
 295                 if (unlikely(error)) {
 296                         kfree_skb(skb);
 297                         return;
 298                 }
 299
 300                 if (is_frag && dp->drop_frags) {
 301                         consume_skb(skb);
 302                         stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
 303                         goto out;
 304                 }
 305
 306                 /* Look up flow. */
 307                 flow = flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
 308                 if (unlikely(!flow)) {
 309                         struct dp_upcall_info upcall;
 310
 311                         upcall.cmd = OVS_PACKET_CMD_MISS;
 312                         upcall.key = &key;
 313                         dp_upcall(dp, skb, &upcall);
 314                         stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
 315                         goto out;
 316                 }
 317
 318                 OVS_CB(skb)->flow = flow;
 319         }
 320
 321         stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
 322         flow_used(OVS_CB(skb)->flow, skb);
 323         execute_actions(dp, skb);
 324
 325 out:
 326         /* Update datapath statistics. */
 327         local_bh_disable();
 328         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 329
 330         write_seqcount_begin(&stats->seqlock);
 331         (*(u64 *)((u8 *)stats + stats_counter_off))++;
 332         write_seqcount_end(&stats->seqlock);
 333
 334         local_bh_enable();
 335 }
 336
 337 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 338 {
 339         u16 csum_start, csum_offset;
 340         __wsum csum;
 341
 342         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 343         csum_start -= skb_headroom(skb);
 344
 345         skb_copy_bits(skb, 0, to, csum_start);
 346
 347         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 348                                       skb->len - csum_start, 0);
 349         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 350 }
 351
 352 static struct genl_family dp_packet_genl_family = {
 353         .id = GENL_ID_GENERATE,
 354         .hdrsize = sizeof(struct ovs_header),
 355         .name = OVS_PACKET_FAMILY,
 356         .version = 1,
 357         .maxattr = OVS_PACKET_ATTR_MAX
 358 };
 359
 360 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
 361 {
 362         struct dp_stats_percpu *stats;
 363         u32 pid;
 364         int err;
 365
 366         if (OVS_CB(skb)->flow)
 367                 pid = OVS_CB(skb)->flow->upcall_pid;
 368         else
 369                 pid = OVS_CB(skb)->vport->upcall_pid;
 370
 371         if (pid == 0) {
 372                 err = -ENOTCONN;
 373                 kfree_skb(skb);
 374                 goto err;
 375         }
 376
 377         forward_ip_summed(skb, true);
 378
 379         /* Break apart GSO packets into their component pieces.  Otherwise
 380          * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
 381         if (skb_is_gso(skb)) {
 382                 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 383
 384                 if (IS_ERR(nskb)) {
 385                         kfree_skb(skb);
 386                         err = PTR_ERR(nskb);
 387                         goto err;
 388                 }
 389                 consume_skb(skb);
 390                 skb = nskb;
 391         }
 392
 393         err = queue_userspace_packets(dp, pid, skb, upcall_info);
 394         if (err)
 395                 goto err;
 396
 397         return 0;
 398
 399 err:
 400         local_bh_disable();
 401         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 402
 403         write_seqcount_begin(&stats->seqlock);
 404         stats->n_lost++;
 405         write_seqcount_end(&stats->seqlock);
 406
 407         local_bh_enable();
 408
 409         return err;
 410 }
 411
 412 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
 413  * 'upcall_info'.  There will be only one packet unless we broke up a GSO
 414  * packet.
 415  */
 416 static int queue_userspace_packets(struct datapath *dp, u32 pid,
 417                                    struct sk_buff *skb,
 418                                    const struct dp_upcall_info *upcall_info)
 419 {
 420         int dp_ifindex;
 421         struct sk_buff *nskb;
 422         int err;
 423
 424         dp_ifindex = get_dpifindex(dp);
 425         if (!dp_ifindex) {
 426                 err = -ENODEV;
 427                 nskb = skb->next;
 428                 goto err_kfree_skbs;
 429         }
 430
 431         do {
 432                 struct ovs_header *upcall;
 433                 struct sk_buff *user_skb; /* to be queued to userspace */
 434                 struct nlattr *nla;
 435                 unsigned int len;
 436
 437                 nskb = skb->next;
 438                 skb->next = NULL;
 439
 440                 err = vlan_deaccel_tag(skb);
 441                 if (unlikely(err))
 442                         goto err_kfree_skbs;
 443
 444                 if (nla_attr_size(skb->len) > USHRT_MAX) {
 445                         err = -EFBIG;
 446                         goto err_kfree_skbs;
 447                 }
 448
 449                 len = sizeof(struct ovs_header);
 450                 len += nla_total_size(skb->len);
 451                 len += nla_total_size(FLOW_BUFSIZE);
 452                 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 453                         len += nla_total_size(8);
 454
 455                 user_skb = genlmsg_new(len, GFP_ATOMIC);
 456                 if (!user_skb) {
 457                         err = -ENOMEM;
 458                         goto err_kfree_skbs;
 459                 }
 460
 461                 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
 462                 upcall->dp_ifindex = dp_ifindex;
 463
 464                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 465                 flow_to_nlattrs(upcall_info->key, user_skb);
 466                 nla_nest_end(user_skb, nla);
 467
 468                 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 469                         nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 470                                                 upcall_info->userdata);
 471
 472                 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 473                 if (skb->ip_summed == CHECKSUM_PARTIAL)
 474                         copy_and_csum_skb(skb, nla_data(nla));
 475                 else
 476                         skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 477
 478                 err = genlmsg_unicast(&init_net, user_skb, pid);
 479                 if (err)
 480                         goto err_kfree_skbs;
 481
 482                 consume_skb(skb);
 483                 skb = nskb;
 484         } while (skb);
 485         return 0;
 486
 487 err_kfree_skbs:
 488         kfree_skb(skb);
 489         while ((skb = nskb) != NULL) {
 490                 nskb = skb->next;
 491                 kfree_skb(skb);
 492         }
 493         return err;
 494 }
 495
 496 /* Called with genl_mutex. */
 497 static int flush_flows(int dp_ifindex)
 498 {
 499         struct flow_table *old_table;
 500         struct flow_table *new_table;
 501         struct datapath *dp;
 502
 503         dp = get_dp(dp_ifindex);
 504         if (!dp)
 505                 return -ENODEV;
 506
 507         old_table = get_table_protected(dp);
 508         new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
 509         if (!new_table)
 510                 return -ENOMEM;
 511
 512         rcu_assign_pointer(dp->table, new_table);
 513
 514         flow_tbl_deferred_destroy(old_table);
 515         return 0;
 516 }
 517
 518 static int validate_actions(const struct nlattr *attr, int depth);
 519
 520 static int validate_sample(const struct nlattr *attr, int depth)
 521 {
 522         static const struct nla_policy sample_policy[OVS_SAMPLE_ATTR_MAX + 1] =
 523         {
 524                 [OVS_SAMPLE_ATTR_PROBABILITY] = {.type = NLA_U32 },
 525                 [OVS_SAMPLE_ATTR_ACTIONS] = {.type = NLA_UNSPEC },
 526         };
 527         struct nlattr *a[OVS_SAMPLE_ATTR_MAX + 1];
 528         int error;
 529
 530         error = nla_parse_nested(a, OVS_SAMPLE_ATTR_MAX, attr, sample_policy);
 531         if (error)
 532                 return error;
 533
 534         if (!a[OVS_SAMPLE_ATTR_PROBABILITY])
 535                 return -EINVAL;
 536         if (!a[OVS_SAMPLE_ATTR_ACTIONS])
 537                 return -EINVAL;
 538
 539         return validate_actions(a[OVS_SAMPLE_ATTR_ACTIONS], (depth + 1));
 540 }
 541
 542 static int validate_actions(const struct nlattr *attr, int depth)
 543 {
 544         const struct nlattr *a;
 545         int rem, err;
 546
 547         if (depth >= SAMPLE_ACTION_DEPTH)
 548                 return -EOVERFLOW;
 549
 550         nla_for_each_nested(a, attr, rem) {
 551                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 552                         [OVS_ACTION_ATTR_OUTPUT] = 4,
 553                         [OVS_ACTION_ATTR_USERSPACE] = 8,
 554                         [OVS_ACTION_ATTR_PUSH_VLAN] = 2,
 555                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 556                         [OVS_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
 557                         [OVS_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
 558                         [OVS_ACTION_ATTR_SET_NW_SRC] = 4,
 559                         [OVS_ACTION_ATTR_SET_NW_DST] = 4,
 560                         [OVS_ACTION_ATTR_SET_NW_TOS] = 1,
 561                         [OVS_ACTION_ATTR_SET_TP_SRC] = 2,
 562                         [OVS_ACTION_ATTR_SET_TP_DST] = 2,
 563                         [OVS_ACTION_ATTR_SET_TUNNEL] = 8,
 564                         [OVS_ACTION_ATTR_SET_PRIORITY] = 4,
 565                         [OVS_ACTION_ATTR_POP_PRIORITY] = 0,
 566                 };
 567                 int type = nla_type(a);
 568
 569                 /* Match expected attr len for given attr len except for
 570                  * OVS_ACTION_ATTR_SAMPLE, as it has nested actions list which
 571                  * is variable size. */
 572                 if (type > OVS_ACTION_ATTR_MAX ||
 573                    (nla_len(a) != action_lens[type] &&
 574                           type != OVS_ACTION_ATTR_SAMPLE))
 575                         return -EINVAL;
 576
 577                 switch (type) {
 578                 case OVS_ACTION_ATTR_UNSPEC:
 579                         return -EINVAL;
 580
 581                 case OVS_ACTION_ATTR_USERSPACE:
 582                 case OVS_ACTION_ATTR_POP_VLAN:
 583                 case OVS_ACTION_ATTR_SET_DL_SRC:
 584                 case OVS_ACTION_ATTR_SET_DL_DST:
 585                 case OVS_ACTION_ATTR_SET_NW_SRC:
 586                 case OVS_ACTION_ATTR_SET_NW_DST:
 587                 case OVS_ACTION_ATTR_SET_TP_SRC:
 588                 case OVS_ACTION_ATTR_SET_TP_DST:
 589                 case OVS_ACTION_ATTR_SET_TUNNEL:
 590                 case OVS_ACTION_ATTR_SET_PRIORITY:
 591                 case OVS_ACTION_ATTR_POP_PRIORITY:
 592                         /* No validation needed. */
 593                         break;
 594
 595                 case OVS_ACTION_ATTR_OUTPUT:
 596                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 597                                 return -EINVAL;
 598                         break;
 599
 600                 case OVS_ACTION_ATTR_PUSH_VLAN:
 601                         if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
 602                                 return -EINVAL;
 603                         break;
 604
 605                 case OVS_ACTION_ATTR_SET_NW_TOS:
 606                         if (nla_get_u8(a) & INET_ECN_MASK)
 607                                 return -EINVAL;
 608                         break;
 609
 610                 case OVS_ACTION_ATTR_SAMPLE:
 611                         err = validate_sample(a, depth);
 612                         if (err)
 613                                 return err;
 614                         break;
 615
 616                 default:
 617                         return -EOPNOTSUPP;
 618                 }
 619         }
 620
 621         if (rem > 0)
 622                 return -EINVAL;
 623
 624         return 0;
 625 }
 626 static void clear_stats(struct sw_flow *flow)
 627 {
 628         flow->used = 0;
 629         flow->tcp_flags = 0;
 630         flow->packet_count = 0;
 631         flow->byte_count = 0;
 632 }
 633
 634 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 635 {
 636         struct ovs_header *ovs_header = info->userhdr;
 637         struct nlattr **a = info->attrs;
 638         struct sw_flow_actions *acts;
 639         struct sk_buff *packet;
 640         struct sw_flow *flow;
 641         struct datapath *dp;
 642         struct ethhdr *eth;
 643         bool is_frag;
 644         int len;
 645         int err;
 646         int key_len;
 647
 648         err = -EINVAL;
 649         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 650             !a[OVS_PACKET_ATTR_ACTIONS] ||
 651             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 652                 goto err;
 653
 654         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], 0);
 655         if (err)
 656                 goto err;
 657
 658         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 659         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 660         err = -ENOMEM;
 661         if (!packet)
 662                 goto err;
 663         skb_reserve(packet, NET_IP_ALIGN);
 664
 665         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 666
 667         skb_reset_mac_header(packet);
 668         eth = eth_hdr(packet);
 669
 670         /* Normally, setting the skb 'protocol' field would be handled by a
 671          * call to eth_type_trans(), but it assumes there's a sending
 672          * device, which we may not have. */
 673         if (ntohs(eth->h_proto) >= 1536)
 674                 packet->protocol = eth->h_proto;
 675         else
 676                 packet->protocol = htons(ETH_P_802_2);
 677
 678         /* Build an sw_flow for sending this packet. */
 679         flow = flow_alloc();
 680         err = PTR_ERR(flow);
 681         if (IS_ERR(flow))
 682                 goto err_kfree_skb;
 683
 684         err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
 685         if (err)
 686                 goto err_flow_put;
 687
 688         err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
 689                                          &flow->key.eth.tun_id,
 690                                          a[OVS_PACKET_ATTR_KEY]);
 691         if (err)
 692                 goto err_flow_put;
 693
 694         flow->hash = flow_hash(&flow->key, key_len);
 695
 696         if (a[OVS_PACKET_ATTR_UPCALL_PID])
 697                 flow->upcall_pid = nla_get_u32(a[OVS_PACKET_ATTR_UPCALL_PID]);
 698         else
 699                 flow->upcall_pid = NETLINK_CB(skb).pid;
 700
 701         acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 702         err = PTR_ERR(acts);
 703         if (IS_ERR(acts))
 704                 goto err_flow_put;
 705         rcu_assign_pointer(flow->sf_acts, acts);
 706
 707         OVS_CB(packet)->flow = flow;
 708
 709         rcu_read_lock();
 710         dp = get_dp(ovs_header->dp_ifindex);
 711         err = -ENODEV;
 712         if (!dp)
 713                 goto err_unlock;
 714
 715         if (flow->key.eth.in_port < DP_MAX_PORTS)
 716                 OVS_CB(packet)->vport = get_vport_protected(dp,
 717                                                         flow->key.eth.in_port);
 718
 719         err = execute_actions(dp, packet);
 720         rcu_read_unlock();
 721
 722         flow_put(flow);
 723         return err;
 724
 725 err_unlock:
 726         rcu_read_unlock();
 727 err_flow_put:
 728         flow_put(flow);
 729 err_kfree_skb:
 730         kfree_skb(packet);
 731 err:
 732         return err;
 733 }
 734
 735 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 736         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 737         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 738         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 739         [OVS_PACKET_ATTR_UPCALL_PID] = { .type = NLA_U32 },
 740 };
 741
 742 static struct genl_ops dp_packet_genl_ops[] = {
 743         { .cmd = OVS_PACKET_CMD_EXECUTE,
 744           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 745           .policy = packet_policy,
 746           .doit = ovs_packet_cmd_execute
 747         }
 748 };
 749
 750 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 751 {
 752         int i;
 753         struct flow_table *table = get_table_protected(dp);
 754
 755         stats->n_flows = flow_tbl_count(table);
 756
 757         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 758         for_each_possible_cpu(i) {
 759                 const struct dp_stats_percpu *percpu_stats;
 760                 struct dp_stats_percpu local_stats;
 761                 unsigned seqcount;
 762
 763                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 764
 765                 do {
 766                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 767                         local_stats = *percpu_stats;
 768                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 769
 770                 stats->n_frags += local_stats.n_frags;
 771                 stats->n_hit += local_stats.n_hit;
 772                 stats->n_missed += local_stats.n_missed;
 773                 stats->n_lost += local_stats.n_lost;
 774         }
 775 }
 776
 777 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 778         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 779         [OVS_FLOW_ATTR_UPCALL_PID] = { .type = NLA_U32 },
 780         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 781         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 782 };
 783
 784 static struct genl_family dp_flow_genl_family = {
 785         .id = GENL_ID_GENERATE,
 786         .hdrsize = sizeof(struct ovs_header),
 787         .name = OVS_FLOW_FAMILY,
 788         .version = 1,
 789         .maxattr = OVS_FLOW_ATTR_MAX
 790 };
 791
 792 static struct genl_multicast_group dp_flow_multicast_group = {
 793         .name = OVS_FLOW_MCGROUP
 794 };
 795
 796 /* Called with genl_lock. */
 797 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 798                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 799 {
 800         const int skb_orig_len = skb->len;
 801         const struct sw_flow_actions *sf_acts;
 802         struct ovs_flow_stats stats;
 803         struct ovs_header *ovs_header;
 804         struct nlattr *nla;
 805         unsigned long used;
 806         u8 tcp_flags;
 807         int err;
 808
 809         sf_acts = rcu_dereference_protected(flow->sf_acts,
 810                                             lockdep_genl_is_held());
 811
 812         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 813         if (!ovs_header)
 814                 return -EMSGSIZE;
 815
 816         ovs_header->dp_ifindex = get_dpifindex(dp);
 817
 818         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 819         if (!nla)
 820                 goto nla_put_failure;
 821         err = flow_to_nlattrs(&flow->key, skb);
 822         if (err)
 823                 goto error;
 824         nla_nest_end(skb, nla);
 825
 826         NLA_PUT_U32(skb, OVS_FLOW_ATTR_UPCALL_PID, flow->upcall_pid);
 827
 828         spin_lock_bh(&flow->lock);
 829         used = flow->used;
 830         stats.n_packets = flow->packet_count;
 831         stats.n_bytes = flow->byte_count;
 832         tcp_flags = flow->tcp_flags;
 833         spin_unlock_bh(&flow->lock);
 834
 835         if (used)
 836                 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
 837
 838         if (stats.n_packets)
 839                 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
 840
 841         if (tcp_flags)
 842                 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 843
 844         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 845          * this is the first flow to be dumped into 'skb'.  This is unusual for
 846          * Netlink but individual action lists can be longer than
 847          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 848          * The userspace caller can always fetch the actions separately if it
 849          * really wants them.  (Most userspace callers in fact don't care.)
 850          *
 851          * This can only fail for dump operations because the skb is always
 852          * properly sized for single flows.
 853          */
 854         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 855                       sf_acts->actions);
 856         if (err < 0 && skb_orig_len)
 857                 goto error;
 858
 859         return genlmsg_end(skb, ovs_header);
 860
 861 nla_put_failure:
 862         err = -EMSGSIZE;
 863 error:
 864         genlmsg_cancel(skb, ovs_header);
 865         return err;
 866 }
 867
 868 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 869 {
 870         const struct sw_flow_actions *sf_acts;
 871         int len;
 872
 873         sf_acts = rcu_dereference_protected(flow->sf_acts,
 874                                             lockdep_genl_is_held());
 875
 876         len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
 877         len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 878         len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
 879         len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
 880         len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
 881         return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
 882 }
 883
 884 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 885                                                u32 pid, u32 seq, u8 cmd)
 886 {
 887         struct sk_buff *skb;
 888         int retval;
 889
 890         skb = ovs_flow_cmd_alloc_info(flow);
 891         if (!skb)
 892                 return ERR_PTR(-ENOMEM);
 893
 894         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 895         BUG_ON(retval < 0);
 896         return skb;
 897 }
 898
 899 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 900 {
 901         struct nlattr **a = info->attrs;
 902         struct ovs_header *ovs_header = info->userhdr;
 903         struct sw_flow_key key;
 904         struct sw_flow *flow;
 905         struct sk_buff *reply;
 906         struct datapath *dp;
 907         struct flow_table *table;
 908         int error;
 909         int key_len;
 910
 911         /* Extract key. */
 912         error = -EINVAL;
 913         if (!a[OVS_FLOW_ATTR_KEY])
 914                 goto error;
 915         error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 916         if (error)
 917                 goto error;
 918
 919         /* Validate actions. */
 920         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 921                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], 0);
 922                 if (error)
 923                         goto error;
 924         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 925                 error = -EINVAL;
 926                 goto error;
 927         }
 928
 929         dp = get_dp(ovs_header->dp_ifindex);
 930         error = -ENODEV;
 931         if (!dp)
 932                 goto error;
 933
 934         table = get_table_protected(dp);
 935         flow = flow_tbl_lookup(table, &key, key_len);
 936         if (!flow) {
 937                 struct sw_flow_actions *acts;
 938
 939                 /* Bail out if we're not allowed to create a new flow. */
 940                 error = -ENOENT;
 941                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 942                         goto error;
 943
 944                 /* Expand table, if necessary, to make room. */
 945                 if (flow_tbl_need_to_expand(table)) {
 946                         struct flow_table *new_table;
 947
 948                         new_table = flow_tbl_expand(table);
 949                         if (!IS_ERR(new_table)) {
 950                                 rcu_assign_pointer(dp->table, new_table);
 951                                 flow_tbl_deferred_destroy(table);
 952                                 table = get_table_protected(dp);
 953                         }
 954                 }
 955
 956                 /* Allocate flow. */
 957                 flow = flow_alloc();
 958                 if (IS_ERR(flow)) {
 959                         error = PTR_ERR(flow);
 960                         goto error;
 961                 }
 962                 flow->key = key;
 963                 clear_stats(flow);
 964
 965                 if (a[OVS_FLOW_ATTR_UPCALL_PID])
 966                         flow->upcall_pid = nla_get_u32(a[OVS_FLOW_ATTR_UPCALL_PID]);
 967                 else
 968                         flow->upcall_pid = NETLINK_CB(skb).pid;
 969
 970                 /* Obtain actions. */
 971                 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
 972                 error = PTR_ERR(acts);
 973                 if (IS_ERR(acts))
 974                         goto error_free_flow;
 975                 rcu_assign_pointer(flow->sf_acts, acts);
 976
 977                 /* Put flow in bucket. */
 978                 flow->hash = flow_hash(&key, key_len);
 979                 flow_tbl_insert(table, flow);
 980
 981                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
 982                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
 983         } else {
 984                 /* We found a matching flow. */
 985                 struct sw_flow_actions *old_acts;
 986
 987                 /* Bail out if we're not allowed to modify an existing flow.
 988                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 989                  * because Generic Netlink treats the latter as a dump
 990                  * request.  We also accept NLM_F_EXCL in case that bug ever
 991                  * gets fixed.
 992                  */
 993                 error = -EEXIST;
 994                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
 995                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
 996                         goto error;
 997
 998                 /* Update actions. */
 999                 old_acts = rcu_dereference_protected(flow->sf_acts,
1000                                                      lockdep_genl_is_held());
1001                 if (a[OVS_FLOW_ATTR_ACTIONS] &&
1002                     (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
1003                      memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
1004                             old_acts->actions_len))) {
1005                         struct sw_flow_actions *new_acts;
1006
1007                         new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1008                         error = PTR_ERR(new_acts);
1009                         if (IS_ERR(new_acts))
1010                                 goto error;
1011
1012                         rcu_assign_pointer(flow->sf_acts, new_acts);
1013                         flow_deferred_free_acts(old_acts);
1014                 }
1015
1016                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1017                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1018
1019                 if (a[OVS_FLOW_ATTR_UPCALL_PID])
1020                         flow->upcall_pid = nla_get_u32(a[OVS_FLOW_ATTR_UPCALL_PID]);
1021
1022                 /* Clear stats. */
1023                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1024                         spin_lock_bh(&flow->lock);
1025                         clear_stats(flow);
1026                         spin_unlock_bh(&flow->lock);
1027                 }
1028         }
1029
1030         if (!IS_ERR(reply))
1031                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1032                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1033         else
1034                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1035                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1036         return 0;
1037
1038 error_free_flow:
1039         flow_put(flow);
1040 error:
1041         return error;
1042 }
1043
1044 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1045 {
1046         struct nlattr **a = info->attrs;
1047         struct ovs_header *ovs_header = info->userhdr;
1048         struct sw_flow_key key;
1049         struct sk_buff *reply;
1050         struct sw_flow *flow;
1051         struct datapath *dp;
1052         struct flow_table *table;
1053         int err;
1054         int key_len;
1055
1056         if (!a[OVS_FLOW_ATTR_KEY])
1057                 return -EINVAL;
1058         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1059         if (err)
1060                 return err;
1061
1062         dp = get_dp(ovs_header->dp_ifindex);
1063         if (!dp)
1064                 return -ENODEV;
1065
1066         table = get_table_protected(dp);
1067         flow = flow_tbl_lookup(table, &key, key_len);
1068         if (!flow)
1069                 return -ENOENT;
1070
1071         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
1072         if (IS_ERR(reply))
1073                 return PTR_ERR(reply);
1074
1075         return genlmsg_reply(reply, info);
1076 }
1077
1078 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1079 {
1080         struct nlattr **a = info->attrs;
1081         struct ovs_header *ovs_header = info->userhdr;
1082         struct sw_flow_key key;
1083         struct sk_buff *reply;
1084         struct sw_flow *flow;
1085         struct datapath *dp;
1086         struct flow_table *table;
1087         int err;
1088         int key_len;
1089
1090         if (!a[OVS_FLOW_ATTR_KEY])
1091                 return flush_flows(ovs_header->dp_ifindex);
1092         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1093         if (err)
1094                 return err;
1095
1096         dp = get_dp(ovs_header->dp_ifindex);
1097         if (!dp)
1098                 return -ENODEV;
1099
1100         table = get_table_protected(dp);
1101         flow = flow_tbl_lookup(table, &key, key_len);
1102         if (!flow)
1103                 return -ENOENT;
1104
1105         reply = ovs_flow_cmd_alloc_info(flow);
1106         if (!reply)
1107                 return -ENOMEM;
1108
1109         flow_tbl_remove(table, flow);
1110
1111         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1112                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1113         BUG_ON(err < 0);
1114
1115         flow_deferred_free(flow);
1116
1117         genl_notify(reply, genl_info_net(info), info->snd_pid,
1118                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1119         return 0;
1120 }
1121
1122 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1123 {
1124         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1125         struct datapath *dp;
1126
1127         dp = get_dp(ovs_header->dp_ifindex);
1128         if (!dp)
1129                 return -ENODEV;
1130
1131         for (;;) {
1132                 struct sw_flow *flow;
1133                 u32 bucket, obj;
1134
1135                 bucket = cb->args[0];
1136                 obj = cb->args[1];
1137                 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1138                 if (!flow)
1139                         break;
1140
1141                 if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1142                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1143                                            OVS_FLOW_CMD_NEW) < 0)
1144                         break;
1145
1146                 cb->args[0] = bucket;
1147                 cb->args[1] = obj;
1148         }
1149         return skb->len;
1150 }
1151
1152 static struct genl_ops dp_flow_genl_ops[] = {
1153         { .cmd = OVS_FLOW_CMD_NEW,
1154           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1155           .policy = flow_policy,
1156           .doit = ovs_flow_cmd_new_or_set
1157         },
1158         { .cmd = OVS_FLOW_CMD_DEL,
1159           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1160           .policy = flow_policy,
1161           .doit = ovs_flow_cmd_del
1162         },
1163         { .cmd = OVS_FLOW_CMD_GET,
1164           .flags = 0,               /* OK for unprivileged users. */
1165           .policy = flow_policy,
1166           .doit = ovs_flow_cmd_get,
1167           .dumpit = ovs_flow_cmd_dump
1168         },
1169         { .cmd = OVS_FLOW_CMD_SET,
1170           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1171           .policy = flow_policy,
1172           .doit = ovs_flow_cmd_new_or_set,
1173         },
1174 };
1175
1176 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1177 #ifdef HAVE_NLA_NUL_STRING
1178         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1179 #endif
1180         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1181         [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1182 };
1183
1184 static struct genl_family dp_datapath_genl_family = {
1185         .id = GENL_ID_GENERATE,
1186         .hdrsize = sizeof(struct ovs_header),
1187         .name = OVS_DATAPATH_FAMILY,
1188         .version = 1,
1189         .maxattr = OVS_DP_ATTR_MAX
1190 };
1191
1192 static struct genl_multicast_group dp_datapath_multicast_group = {
1193         .name = OVS_DATAPATH_MCGROUP
1194 };
1195
1196 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1197                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1198 {
1199         struct ovs_header *ovs_header;
1200         struct nlattr *nla;
1201         int err;
1202
1203         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1204                                    flags, cmd);
1205         if (!ovs_header)
1206                 goto error;
1207
1208         ovs_header->dp_ifindex = get_dpifindex(dp);
1209
1210         rcu_read_lock();
1211         err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1212         rcu_read_unlock();
1213         if (err)
1214                 goto nla_put_failure;
1215
1216         nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1217         if (!nla)
1218                 goto nla_put_failure;
1219         get_dp_stats(dp, nla_data(nla));
1220
1221         NLA_PUT_U32(skb, OVS_DP_ATTR_IPV4_FRAGS,
1222                     dp->drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO);
1223
1224         return genlmsg_end(skb, ovs_header);
1225
1226 nla_put_failure:
1227         genlmsg_cancel(skb, ovs_header);
1228 error:
1229         return -EMSGSIZE;
1230 }
1231
1232 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1233                                              u32 seq, u8 cmd)
1234 {
1235         struct sk_buff *skb;
1236         int retval;
1237
1238         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1239         if (!skb)
1240                 return ERR_PTR(-ENOMEM);
1241
1242         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1243         if (retval < 0) {
1244                 kfree_skb(skb);
1245                 return ERR_PTR(retval);
1246         }
1247         return skb;
1248 }
1249
1250 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1251 {
1252         if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
1253                 u32 frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
1254
1255                 if (frags != OVS_DP_FRAG_ZERO && frags != OVS_DP_FRAG_DROP)
1256                         return -EINVAL;
1257         }
1258
1259         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1260 }
1261
1262 /* Called with genl_mutex and optionally with RTNL lock also. */
1263 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1264 {
1265         struct datapath *dp;
1266
1267         if (!a[OVS_DP_ATTR_NAME])
1268                 dp = get_dp(ovs_header->dp_ifindex);
1269         else {
1270                 struct vport *vport;
1271
1272                 rcu_read_lock();
1273                 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1274                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1275                 rcu_read_unlock();
1276         }
1277         return dp ? dp : ERR_PTR(-ENODEV);
1278 }
1279
1280 /* Called with genl_mutex. */
1281 static void change_datapath(struct datapath *dp, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1282 {
1283         if (a[OVS_DP_ATTR_IPV4_FRAGS])
1284                 dp->drop_frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]) == OVS_DP_FRAG_DROP;
1285 }
1286
1287 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1288 {
1289         struct nlattr **a = info->attrs;
1290         struct vport_parms parms;
1291         struct sk_buff *reply;
1292         struct datapath *dp;
1293         struct vport *vport;
1294         int err;
1295
1296         err = -EINVAL;
1297         if (!a[OVS_DP_ATTR_NAME])
1298                 goto err;
1299
1300         err = ovs_dp_cmd_validate(a);
1301         if (err)
1302                 goto err;
1303
1304         rtnl_lock();
1305         err = -ENODEV;
1306         if (!try_module_get(THIS_MODULE))
1307                 goto err_unlock_rtnl;
1308
1309         err = -ENOMEM;
1310         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1311         if (dp == NULL)
1312                 goto err_put_module;
1313         INIT_LIST_HEAD(&dp->port_list);
1314
1315         /* Initialize kobject for bridge.  This will be added as
1316          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1317         dp->ifobj.kset = NULL;
1318         kobject_init(&dp->ifobj, &dp_ktype);
1319
1320         /* Allocate table. */
1321         err = -ENOMEM;
1322         rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1323         if (!dp->table)
1324                 goto err_free_dp;
1325
1326         dp->drop_frags = 0;
1327         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1328         if (!dp->stats_percpu) {
1329                 err = -ENOMEM;
1330                 goto err_destroy_table;
1331         }
1332
1333         change_datapath(dp, a);
1334
1335         /* Set up our datapath device. */
1336         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1337         parms.type = OVS_VPORT_TYPE_INTERNAL;
1338         parms.options = NULL;
1339         parms.dp = dp;
1340         parms.port_no = OVSP_LOCAL;
1341         if (a[OVS_DP_ATTR_UPCALL_PID])
1342                 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1343         else
1344                 parms.upcall_pid = NETLINK_CB(skb).pid;
1345
1346         vport = new_vport(&parms);
1347         if (IS_ERR(vport)) {
1348                 err = PTR_ERR(vport);
1349                 if (err == -EBUSY)
1350                         err = -EEXIST;
1351
1352                 goto err_destroy_percpu;
1353         }
1354
1355         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1356         err = PTR_ERR(reply);
1357         if (IS_ERR(reply))
1358                 goto err_destroy_local_port;
1359
1360         list_add_tail(&dp->list_node, &dps);
1361         dp_sysfs_add_dp(dp);
1362
1363         rtnl_unlock();
1364
1365         genl_notify(reply, genl_info_net(info), info->snd_pid,
1366                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1367         return 0;
1368
1369 err_destroy_local_port:
1370         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1371 err_destroy_percpu:
1372         free_percpu(dp->stats_percpu);
1373 err_destroy_table:
1374         flow_tbl_destroy(get_table_protected(dp));
1375 err_free_dp:
1376         kfree(dp);
1377 err_put_module:
1378         module_put(THIS_MODULE);
1379 err_unlock_rtnl:
1380         rtnl_unlock();
1381 err:
1382         return err;
1383 }
1384
1385 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1386 {
1387         struct vport *vport, *next_vport;
1388         struct sk_buff *reply;
1389         struct datapath *dp;
1390         int err;
1391
1392         err = ovs_dp_cmd_validate(info->attrs);
1393         if (err)
1394                 goto exit;
1395
1396         rtnl_lock();
1397         dp = lookup_datapath(info->userhdr, info->attrs);
1398         err = PTR_ERR(dp);
1399         if (IS_ERR(dp))
1400                 goto exit_unlock;
1401
1402         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
1403         err = PTR_ERR(reply);
1404         if (IS_ERR(reply))
1405                 goto exit_unlock;
1406
1407         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1408                 if (vport->port_no != OVSP_LOCAL)
1409                         dp_detach_port(vport);
1410
1411         dp_sysfs_del_dp(dp);
1412         list_del(&dp->list_node);
1413         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1414
1415         /* rtnl_unlock() will wait until all the references to devices that
1416          * are pending unregistration have been dropped.  We do it here to
1417          * ensure that any internal devices (which contain DP pointers) are
1418          * fully destroyed before freeing the datapath.
1419          */
1420         rtnl_unlock();
1421
1422         call_rcu(&dp->rcu, destroy_dp_rcu);
1423         module_put(THIS_MODULE);
1424
1425         genl_notify(reply, genl_info_net(info), info->snd_pid,
1426                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1427
1428         return 0;
1429
1430 exit_unlock:
1431         rtnl_unlock();
1432 exit:
1433         return err;
1434 }
1435
1436 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1437 {
1438         struct sk_buff *reply;
1439         struct datapath *dp;
1440         int err;
1441
1442         err = ovs_dp_cmd_validate(info->attrs);
1443         if (err)
1444                 return err;
1445
1446         dp = lookup_datapath(info->userhdr, info->attrs);
1447         if (IS_ERR(dp))
1448                 return PTR_ERR(dp);
1449
1450         change_datapath(dp, info->attrs);
1451
1452         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1453         if (IS_ERR(reply)) {
1454                 err = PTR_ERR(reply);
1455                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1456                                 dp_datapath_multicast_group.id, err);
1457                 return 0;
1458         }
1459
1460         genl_notify(reply, genl_info_net(info), info->snd_pid,
1461                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1462         return 0;
1463 }
1464
1465 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1466 {
1467         struct sk_buff *reply;
1468         struct datapath *dp;
1469         int err;
1470
1471         err = ovs_dp_cmd_validate(info->attrs);
1472         if (err)
1473                 return err;
1474
1475         dp = lookup_datapath(info->userhdr, info->attrs);
1476         if (IS_ERR(dp))
1477                 return PTR_ERR(dp);
1478
1479         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1480         if (IS_ERR(reply))
1481                 return PTR_ERR(reply);
1482
1483         return genlmsg_reply(reply, info);
1484 }
1485
1486 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1487 {
1488         struct datapath *dp;
1489         int skip = cb->args[0];
1490         int i = 0;
1491
1492         list_for_each_entry (dp, &dps, list_node) {
1493                 if (i < skip)
1494                         continue;
1495                 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1496                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1497                                          OVS_DP_CMD_NEW) < 0)
1498                         break;
1499                 i++;
1500         }
1501
1502         cb->args[0] = i;
1503
1504         return skb->len;
1505 }
1506
1507 static struct genl_ops dp_datapath_genl_ops[] = {
1508         { .cmd = OVS_DP_CMD_NEW,
1509           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1510           .policy = datapath_policy,
1511           .doit = ovs_dp_cmd_new
1512         },
1513         { .cmd = OVS_DP_CMD_DEL,
1514           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1515           .policy = datapath_policy,
1516           .doit = ovs_dp_cmd_del
1517         },
1518         { .cmd = OVS_DP_CMD_GET,
1519           .flags = 0,               /* OK for unprivileged users. */
1520           .policy = datapath_policy,
1521           .doit = ovs_dp_cmd_get,
1522           .dumpit = ovs_dp_cmd_dump
1523         },
1524         { .cmd = OVS_DP_CMD_SET,
1525           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1526           .policy = datapath_policy,
1527           .doit = ovs_dp_cmd_set,
1528         },
1529 };
1530
1531 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1532 #ifdef HAVE_NLA_NUL_STRING
1533         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1534         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1535         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1536 #else
1537         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1538         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1539 #endif
1540         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1541         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1542         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1543         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1544 };
1545
1546 static struct genl_family dp_vport_genl_family = {
1547         .id = GENL_ID_GENERATE,
1548         .hdrsize = sizeof(struct ovs_header),
1549         .name = OVS_VPORT_FAMILY,
1550         .version = 1,
1551         .maxattr = OVS_VPORT_ATTR_MAX
1552 };
1553
1554 struct genl_multicast_group dp_vport_multicast_group = {
1555         .name = OVS_VPORT_MCGROUP
1556 };
1557
1558 /* Called with RTNL lock or RCU read lock. */
1559 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1560                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1561 {
1562         struct ovs_header *ovs_header;
1563         struct nlattr *nla;
1564         int ifindex;
1565         int err;
1566
1567         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1568                                  flags, cmd);
1569         if (!ovs_header)
1570                 return -EMSGSIZE;
1571
1572         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1573
1574         NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1575         NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1576         NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1577         NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1578
1579         nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats));
1580         if (!nla)
1581                 goto nla_put_failure;
1582
1583         vport_get_stats(vport, nla_data(nla));
1584
1585         NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1586
1587         err = vport_get_options(vport, skb);
1588         if (err == -EMSGSIZE)
1589                 goto error;
1590
1591         ifindex = vport_get_ifindex(vport);
1592         if (ifindex > 0)
1593                 NLA_PUT_U32(skb, OVS_VPORT_ATTR_IFINDEX, ifindex);
1594
1595         return genlmsg_end(skb, ovs_header);
1596
1597 nla_put_failure:
1598         err = -EMSGSIZE;
1599 error:
1600         genlmsg_cancel(skb, ovs_header);
1601         return err;
1602 }
1603
1604 /* Called with RTNL lock or RCU read lock. */
1605 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1606                                          u32 seq, u8 cmd)
1607 {
1608         struct sk_buff *skb;
1609         int retval;
1610
1611         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1612         if (!skb)
1613                 return ERR_PTR(-ENOMEM);
1614
1615         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1616         if (retval < 0) {
1617                 kfree_skb(skb);
1618                 return ERR_PTR(retval);
1619         }
1620         return skb;
1621 }
1622
1623 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1624 {
1625         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1626 }
1627
1628 /* Called with RTNL lock or RCU read lock. */
1629 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1630                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1631 {
1632         struct datapath *dp;
1633         struct vport *vport;
1634
1635         if (a[OVS_VPORT_ATTR_NAME]) {
1636                 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1637                 if (!vport)
1638                         return ERR_PTR(-ENODEV);
1639                 return vport;
1640         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1641                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1642
1643                 if (port_no >= DP_MAX_PORTS)
1644                         return ERR_PTR(-EFBIG);
1645
1646                 dp = get_dp(ovs_header->dp_ifindex);
1647                 if (!dp)
1648                         return ERR_PTR(-ENODEV);
1649
1650                 vport = get_vport_protected(dp, port_no);
1651                 if (!vport)
1652                         return ERR_PTR(-ENOENT);
1653                 return vport;
1654         } else
1655                 return ERR_PTR(-EINVAL);
1656 }
1657
1658 /* Called with RTNL lock. */
1659 static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1660 {
1661         int err = 0;
1662
1663         if (a[OVS_VPORT_ATTR_STATS])
1664                 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1665
1666         if (a[OVS_VPORT_ATTR_ADDRESS])
1667                 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1668
1669         return err;
1670 }
1671
1672 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1673 {
1674         struct nlattr **a = info->attrs;
1675         struct ovs_header *ovs_header = info->userhdr;
1676         struct vport_parms parms;
1677         struct sk_buff *reply;
1678         struct vport *vport;
1679         struct datapath *dp;
1680         u32 port_no;
1681         int err;
1682
1683         err = -EINVAL;
1684         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE])
1685                 goto exit;
1686
1687         err = ovs_vport_cmd_validate(a);
1688         if (err)
1689                 goto exit;
1690
1691         rtnl_lock();
1692         dp = get_dp(ovs_header->dp_ifindex);
1693         err = -ENODEV;
1694         if (!dp)
1695                 goto exit_unlock;
1696
1697         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1698                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1699
1700                 err = -EFBIG;
1701                 if (port_no >= DP_MAX_PORTS)
1702                         goto exit_unlock;
1703
1704                 vport = get_vport_protected(dp, port_no);
1705                 err = -EBUSY;
1706                 if (vport)
1707                         goto exit_unlock;
1708         } else {
1709                 for (port_no = 1; ; port_no++) {
1710                         if (port_no >= DP_MAX_PORTS) {
1711                                 err = -EFBIG;
1712                                 goto exit_unlock;
1713                         }
1714                         vport = get_vport_protected(dp, port_no);
1715                         if (!vport)
1716                                 break;
1717                 }
1718         }
1719
1720         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1721         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1722         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1723         parms.dp = dp;
1724         parms.port_no = port_no;
1725         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1726                 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1727         else
1728                 parms.upcall_pid = NETLINK_CB(skb).pid;
1729
1730         vport = new_vport(&parms);
1731         err = PTR_ERR(vport);
1732         if (IS_ERR(vport))
1733                 goto exit_unlock;
1734
1735         dp_sysfs_add_if(vport);
1736
1737         err = change_vport(vport, a);
1738         if (!err) {
1739                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1740                                                  info->snd_seq, OVS_VPORT_CMD_NEW);
1741                 if (IS_ERR(reply))
1742                         err = PTR_ERR(reply);
1743         }
1744         if (err) {
1745                 dp_detach_port(vport);
1746                 goto exit_unlock;
1747         }
1748         genl_notify(reply, genl_info_net(info), info->snd_pid,
1749                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1750
1751
1752 exit_unlock:
1753         rtnl_unlock();
1754 exit:
1755         return err;
1756 }
1757
1758 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1759 {
1760         struct nlattr **a = info->attrs;
1761         struct sk_buff *reply;
1762         struct vport *vport;
1763         int err;
1764
1765         err = ovs_vport_cmd_validate(a);
1766         if (err)
1767                 goto exit;
1768
1769         rtnl_lock();
1770         vport = lookup_vport(info->userhdr, a);
1771         err = PTR_ERR(vport);
1772         if (IS_ERR(vport))
1773                 goto exit_unlock;
1774
1775         err = 0;
1776         if (a[OVS_VPORT_ATTR_OPTIONS])
1777                 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1778         if (!err)
1779                 err = change_vport(vport, a);
1780         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1781                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1782
1783         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1784                                          OVS_VPORT_CMD_NEW);
1785         if (IS_ERR(reply)) {
1786                 err = PTR_ERR(reply);
1787                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1788                                 dp_vport_multicast_group.id, err);
1789                 return 0;
1790         }
1791
1792         genl_notify(reply, genl_info_net(info), info->snd_pid,
1793                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1794
1795 exit_unlock:
1796         rtnl_unlock();
1797 exit:
1798         return err;
1799 }
1800
1801 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1802 {
1803         struct nlattr **a = info->attrs;
1804         struct sk_buff *reply;
1805         struct vport *vport;
1806         int err;
1807
1808         err = ovs_vport_cmd_validate(a);
1809         if (err)
1810                 goto exit;
1811
1812         rtnl_lock();
1813         vport = lookup_vport(info->userhdr, a);
1814         err = PTR_ERR(vport);
1815         if (IS_ERR(vport))
1816                 goto exit_unlock;
1817
1818         if (vport->port_no == OVSP_LOCAL) {
1819                 err = -EINVAL;
1820                 goto exit_unlock;
1821         }
1822
1823         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1824                                          OVS_VPORT_CMD_DEL);
1825         err = PTR_ERR(reply);
1826         if (IS_ERR(reply))
1827                 goto exit_unlock;
1828
1829         dp_detach_port(vport);
1830
1831         genl_notify(reply, genl_info_net(info), info->snd_pid,
1832                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1833
1834 exit_unlock:
1835         rtnl_unlock();
1836 exit:
1837         return err;
1838 }
1839
1840 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1841 {
1842         struct nlattr **a = info->attrs;
1843         struct ovs_header *ovs_header = info->userhdr;
1844         struct sk_buff *reply;
1845         struct vport *vport;
1846         int err;
1847
1848         err = ovs_vport_cmd_validate(a);
1849         if (err)
1850                 goto exit;
1851
1852         rcu_read_lock();
1853         vport = lookup_vport(ovs_header, a);
1854         err = PTR_ERR(vport);
1855         if (IS_ERR(vport))
1856                 goto exit_unlock;
1857
1858         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1859                                          OVS_VPORT_CMD_NEW);
1860         err = PTR_ERR(reply);
1861         if (IS_ERR(reply))
1862                 goto exit_unlock;
1863
1864         rcu_read_unlock();
1865
1866         return genlmsg_reply(reply, info);
1867
1868 exit_unlock:
1869         rcu_read_unlock();
1870 exit:
1871         return err;
1872 }
1873
1874 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1875 {
1876         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1877         struct datapath *dp;
1878         u32 port_no;
1879         int retval;
1880
1881         dp = get_dp(ovs_header->dp_ifindex);
1882         if (!dp)
1883                 return -ENODEV;
1884
1885         rcu_read_lock();
1886         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1887                 struct vport *vport;
1888
1889                 vport = get_vport_protected(dp, port_no);
1890                 if (!vport)
1891                         continue;
1892
1893                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1894                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1895                                             OVS_VPORT_CMD_NEW) < 0)
1896                         break;
1897         }
1898         rcu_read_unlock();
1899
1900         cb->args[0] = port_no;
1901         retval = skb->len;
1902
1903         return retval;
1904 }
1905
1906 static struct genl_ops dp_vport_genl_ops[] = {
1907         { .cmd = OVS_VPORT_CMD_NEW,
1908           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1909           .policy = vport_policy,
1910           .doit = ovs_vport_cmd_new
1911         },
1912         { .cmd = OVS_VPORT_CMD_DEL,
1913           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1914           .policy = vport_policy,
1915           .doit = ovs_vport_cmd_del
1916         },
1917         { .cmd = OVS_VPORT_CMD_GET,
1918           .flags = 0,               /* OK for unprivileged users. */
1919           .policy = vport_policy,
1920           .doit = ovs_vport_cmd_get,
1921           .dumpit = ovs_vport_cmd_dump
1922         },
1923         { .cmd = OVS_VPORT_CMD_SET,
1924           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1925           .policy = vport_policy,
1926           .doit = ovs_vport_cmd_set,
1927         },
1928 };
1929
1930 struct genl_family_and_ops {
1931         struct genl_family *family;
1932         struct genl_ops *ops;
1933         int n_ops;
1934         struct genl_multicast_group *group;
1935 };
1936
1937 static const struct genl_family_and_ops dp_genl_families[] = {
1938         { &dp_datapath_genl_family,
1939           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1940           &dp_datapath_multicast_group },
1941         { &dp_vport_genl_family,
1942           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1943           &dp_vport_multicast_group },
1944         { &dp_flow_genl_family,
1945           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1946           &dp_flow_multicast_group },
1947         { &dp_packet_genl_family,
1948           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1949           NULL },
1950 };
1951
1952 static void dp_unregister_genl(int n_families)
1953 {
1954         int i;
1955
1956         for (i = 0; i < n_families; i++)
1957                 genl_unregister_family(dp_genl_families[i].family);
1958 }
1959
1960 static int dp_register_genl(void)
1961 {
1962         int n_registered;
1963         int err;
1964         int i;
1965
1966         n_registered = 0;
1967         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1968                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1969
1970                 err = genl_register_family_with_ops(f->family, f->ops,
1971                                                     f->n_ops);
1972                 if (err)
1973                         goto error;
1974                 n_registered++;
1975
1976                 if (f->group) {
1977                         err = genl_register_mc_group(f->family, f->group);
1978                         if (err)
1979                                 goto error;
1980                 }
1981         }
1982
1983         return 0;
1984
1985 error:
1986         dp_unregister_genl(n_registered);
1987         return err;
1988 }
1989
1990 static int __init dp_init(void)
1991 {
1992         struct sk_buff *dummy_skb;
1993         int err;
1994
1995         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1996
1997         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
1998
1999         err = tnl_init();
2000         if (err)
2001                 goto error;
2002
2003         err = flow_init();
2004         if (err)
2005                 goto error_tnl_exit;
2006
2007         err = vport_init();
2008         if (err)
2009                 goto error_flow_exit;
2010
2011         err = register_netdevice_notifier(&dp_device_notifier);
2012         if (err)
2013                 goto error_vport_exit;
2014
2015         err = dp_register_genl();
2016         if (err < 0)
2017                 goto error_unreg_notifier;
2018
2019         return 0;
2020
2021 error_unreg_notifier:
2022         unregister_netdevice_notifier(&dp_device_notifier);
2023 error_vport_exit:
2024         vport_exit();
2025 error_flow_exit:
2026         flow_exit();
2027 error_tnl_exit:
2028         tnl_exit();
2029 error:
2030         return err;
2031 }
2032
2033 static void dp_cleanup(void)
2034 {
2035         rcu_barrier();
2036         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2037         unregister_netdevice_notifier(&dp_device_notifier);
2038         vport_exit();
2039         flow_exit();
2040         tnl_exit();
2041 }
2042
2043 module_init(dp_init);
2044 module_exit(dp_cleanup);
2045
2046 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2047 MODULE_LICENSE("GPL");